From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner+w=401wt.eu-S1754842AbZIKRAo@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
	id S1754842AbZIKRAo (ORCPT <rfc822;w@1wt.eu>);
	Fri, 11 Sep 2009 13:00:44 -0400
Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1754827AbZIKRAk
	(ORCPT <rfc822;linux-kernel-outgoing>);
	Fri, 11 Sep 2009 13:00:40 -0400
Received: from casper.infradead.org ([85.118.1.10]:55900 "EHLO
	casper.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1754808AbZIKRAi (ORCPT
	<rfc822;linux-kernel@vger.kernel.org>);
	Fri, 11 Sep 2009 13:00:38 -0400
Subject: Re: [RFC][v6][PATCH 0/9] clone_with_pids() syscall
From: Peter Zijlstra <peterz@infradead.org>
To: Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>, linux-kernel@vger.kernel.org,
       Oren Laadan <orenl@cs.columbia.edu>,
       "Eric W. Biederman" <ebiederm@xmission.com>,
       Alexey Dobriyan <adobriyan@gmail.com>,
       Pavel Emelyanov <xemul@openvz.org>, Andrew Morton <akpm@osdl.org>,
       torvalds@linux-foundation.org, mikew@google.com, mingo@elte.hu,
       hpa@zytor.com, Nathan Lynch <nathanl@austin.ibm.com>,
       container@us.ibm.com, sukadev@us.ibm.com
In-Reply-To: <20090911164703.GA29169@us.ibm.com>
References: <20090910060627.GA24343@us.ibm.com>
	 <1252668148.7126.13.camel@laptop> <200909111334.45241.arnd@arndb.de>
	 <20090911164703.GA29169@us.ibm.com>
Content-Type: text/plain
Date: Fri, 11 Sep 2009 19:00:04 +0200
Message-Id: <1252688404.7126.40.camel@laptop>
Mime-Version: 1.0
X-Mailer: Evolution 2.26.1 
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Fri, 2009-09-11 at 09:47 -0700, Sukadev Bhattiprolu wrote:
> | Also, if you're passing a struct, why not put nr_pids in there, and
> | replace clone_pid_struct with a simple array? That would give us
> | 
> | struct clone_struct {
> |         u64 flags;
> |         u64 child_stack;
> |         u32 child_tid;
> |         u32 parent_tid;
> | 	u32 nr_pids;
> | 	u32 reserved1;
> | 	u64 reserved2;
> | };
> | 
> | int clone2(struct clone_struct *cs, pid_t *pids);
> 
> My only concern with this approach was the extra copy_from_user() in the
> common case (i.e when not using the extended features).  I assume the
> overhead of copy_from_user() is small enough to be ignored ?

I would think so, esp for small structure, it would be a cache hot copy.
That is, for x86 I doubt it'll show up. No idea what hoops other arch
have to jump through to get copy_from_user() doing what it does.